High-resolution agricultural drought hazard mapping using the potential of geospatial data and machine learning...

Area of Science:

• Environmental Science
• Agricultural Science
• Remote Sensing
• Machine Learning

Background:

• Effective Agricultural Drought Hazard (ADH) zone delineation is vital for water security and mitigating crop losses in semi-arid regions.
• Conventional drought assessment methods fail to capture complex interactions of geo-environmental drivers in rainfed basins.

Purpose of the Study:

• To develop and evaluate a Machine Learning (ML)-geospatial framework for improved ADH assessment.
• To integrate satellite-derived indices and soil-hydrological parameters for non-linear drought driver analysis.

Main Methods:

• Utilized four ML models: Random Forest (RF), Artificial Neural Network (ANN), Support Vector Machine (SVM), and Adaptive Regression (AR).
• Incorporated eight geo-environmental input variables for drought modeling.
• Evaluated model performance using AUC-ROC and RMSE in the Upper Dwarakeshwar River Basin (UDRB).

Main Results:

• The RF model achieved the highest performance (97.8% AUC-ROC, 0.26 RMSE), followed by SVM (94.6%, 0.28) and ANN (93.8%, 0.32).
• ADH mapping revealed 24.85-44.35% of UDRB as very high and 16.96-22.86% as high ADH regions.
• Significant portions of the drought-prone UDRB require targeted drought mitigation strategies.

Conclusions:

• The ML-geospatial framework effectively delineates ADH zones, outperforming conventional methods.
• Findings support early drought warning, emergency preparedness, and precision agriculture in rainfed basins.
• This approach enhances resilience for climate-vulnerable farming communities dependent on agriculture.